A Learning Method based on DBN for Establishing the Correspondence between Abnormal Point Sets

Fang Li, Fangfang Lu, Xinrong Chen


The establishment of the right correspondence is one of the most basic and the most critical technology in point set registration. For the damaged data of point set in the existence of outliers, noise or missing points, it is difficult to distinguish the abnormal points from the normal and the correspondence between the point sets are affected by these abnormal points. Based on the prior that there is some relation or distinction between the normal points and the abnormal points, we formulate the estimation of correspondence problem by the machine learning procedure. In this paper, considering the feature of error between two point sets, a Deep Belief Networks (DBN) based learning method is proposed to train the networks with the normal point sets. Using the trained DBN, the outlier and unmatched points can be identified. Experiments confirm that our method can well detect noise in the point sets. Even in the case of missing data, our method can identify almost all matched points.


Full Text:



  • There are currently no refbacks.